remove extra libtorch dependency (#14928)

This PR was created by the merge bot to help merge the original PR into
the main branch.
ghstack PR number: #14919 by
@Gasoonjia
^ Please use this as the source of truth for the PR details, comments,
and reviews
ghstack PR base:
https://github.com/pytorch/executorch/tree/gh/gasoonjia/51/base
ghstack PR head:
https://github.com/pytorch/executorch/tree/gh/gasoonjia/51/head
Merge bot PR base:
https://github.com/pytorch/executorch/tree/gh/gasoonjia/50/orig
Merge bot PR head:
https://github.com/pytorch/executorch/tree/gh/gasoonjia/51/orig
Differential Revision:
[D84207378](https://our.internmc.facebook.com/intern/diff/D84207378/)
@diff-train-skip-merge

---------

Co-authored-by: gasoonjia <[email protected]>

Author: pytorchbot

backends/aoti/CMakeLists.txt

@@ -40,7 +40,7 @@ target_compile_options(aoti_common PUBLIC -fexceptions -frtti -fPIC)
 # Ensure symbols are exported properly
 target_link_options(aoti_common PUBLIC -Wl,--export-dynamic)
 
-# Link against PyTorch libraries and standard libraries
+# Link against ExecuTorch libraries and standard libraries
 target_link_libraries(aoti_common PUBLIC extension_tensor ${CMAKE_DL_LIBS})
 executorch_target_link_options_shared_lib(aoti_common)
 

backends/aoti/common_shims.cpp

@@ -127,11 +127,18 @@ int32_t aoti_torch_layout_strided() {
 }
 
 // Dtype constants - these return the PyTorch dtype codes
-// Currently only float32 is supported, but using robust enum-based approach
 int32_t aoti_torch_dtype_float32() {
   return 6; // PyTorch's float32 dtype code
 }
 
+int32_t aoti_torch_dtype_bfloat16() {
+  return 15; // PyTorch's bfloat16 dtype code
+}
+
+int32_t aoti_torch_dtype_int64() {
+  return 4; // PyTorch's int64 dtype code
+}
+
 // Cleanup functions
 void cleanup_tensor_metadata() {
   internal::tensor_to_sizes.clear();

backends/aoti/common_shims.h

@@ -58,6 +58,8 @@ AOTITorchError aoti_torch_get_dim(Tensor* tensor, int64_t* ret_dim);
 int32_t aoti_torch_device_type_cpu();
 int32_t aoti_torch_layout_strided();
 int32_t aoti_torch_dtype_float32();
+int32_t aoti_torch_dtype_bfloat16();
+int32_t aoti_torch_dtype_int64();
 
 // Autograd mode functions
 int32_t aoti_torch_grad_mode_is_enabled();

backends/cuda/CMakeLists.txt

@@ -55,10 +55,7 @@ target_link_options(aoti_cuda PUBLIC -Wl,--export-dynamic)
 
 # Link against CUDA::cudart, common AOTI library, and PyTorch CUDA libraries
 target_link_libraries(
-  aoti_cuda
-  PUBLIC aoti_common CUDA::cudart ${CMAKE_DL_LIBS}
-         # Link PyTorch libraries for AOTI CUDA functions
-         ${TORCH_LIBRARIES}
+  aoti_cuda PUBLIC aoti_common CUDA::cudart ${CMAKE_DL_LIBS}
 )
 # If you need other CUDA libraries, link them similarly:
 # target_link_libraries(aoti_cuda PUBLIC CUDA::cublas CUDA::cufft ...)

backends/cuda/runtime/cuda_backend.cpp

@@ -26,9 +26,7 @@
 #include <executorch/backends/cuda/runtime/shims/memory.h>
 #include <executorch/backends/cuda/runtime/utils.h>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 #define LOAD_SYMBOL(name, handle)                                \
   do {                                                           \
@@ -335,14 +333,13 @@ class ET_EXPERIMENTAL CudaBackend final
   }
 };
 
-} // namespace cuda
+} // namespace executorch::backends::cuda
 
+namespace executorch::backends {
 namespace {
 auto cls = cuda::CudaBackend();
 executorch::runtime::Backend backend{"CudaBackend", &cls};
 static executorch::runtime::Error success_with_compiler =
     register_backend(backend);
 } // namespace
-
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends

backends/cuda/runtime/guard.cpp

@@ -9,9 +9,7 @@
 #include <executorch/backends/cuda/runtime/guard.h>
 #include <executorch/runtime/platform/log.h>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 namespace {
 // Thread-local stream storage (private to this file)
@@ -146,6 +144,4 @@ Result<CUDAStreamGuard> CUDAStreamGuard::create(
   return stream_guard;
 }
 
-} // namespace cuda
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends::cuda

backends/cuda/runtime/guard.h

@@ -14,9 +14,7 @@
 #include <executorch/runtime/core/result.h>
 #include <cstdint>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 using executorch::runtime::Error;
 using executorch::runtime::Result;
@@ -190,6 +188,4 @@ class CUDAStreamGuard {
   DeviceIndex device_index_;
 };
 
-} // namespace cuda
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends::cuda

backends/cuda/runtime/shims/cuda_guard.cpp

@@ -8,9 +8,7 @@
 
 #include <executorch/backends/cuda/runtime/shims/cuda_guard.h>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 extern "C" {
 
@@ -104,6 +102,4 @@ AOTITorchError aoti_torch_get_current_cuda_stream(
 
 } // extern "C"
 
-} // namespace cuda
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends::cuda

backends/cuda/runtime/shims/cuda_guard.h

@@ -13,9 +13,7 @@
 #include <executorch/backends/cuda/runtime/guard.h>
 #include <cstdint>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 using executorch::backends::aoti::AOTITorchError;
 
@@ -99,6 +97,4 @@ AOTITorchError aoti_torch_get_current_cuda_stream(
 
 } // extern "C"
 
-} // namespace cuda
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends::cuda

backends/cuda/runtime/shims/memory.cpp

@@ -19,9 +19,7 @@
 #include <unordered_set>
 #include <vector>
 
-namespace executorch {
-namespace backends {
-namespace cuda {
+namespace executorch::backends::cuda {
 
 using executorch::aten::SizesType;
 using executorch::aten::StridesType;
@@ -659,6 +657,4 @@ AOTITorchError aoti_torch__reinterpret_tensor(
 
 } // extern "C"
 
-} // namespace cuda
-} // namespace backends
-} // namespace executorch
+} // namespace executorch::backends::cuda